Method, device, equipment and medium for predicting disease association of long non-coding RNA

Abstract

This invention belongs to the field of machine learning and discloses a method, apparatus, device, and medium for predicting the association between multiple non-coding RNAs and diseases. The method includes constructing a bioinformatics network among miRNAs, lncRNAs, and diseases; extracting global information of a first node and global information of a second node based on the bioinformatics network; constructing unconnected subgraphs of the first and second nodes; aggregating the features of neighboring nodes of the first and second nodes to obtain local information of the first and second nodes; concatenating the global and local information of the first and second nodes to obtain feature representations of the first and second nodes; and inputting the feature representations of the first and second nodes into a preset association prediction model to obtain the association prediction results of the first and second nodes. This effectively improves the accuracy of the prediction results and enables the analysis of the association between multiple non-coding RNAs and diseases.

Description

Technical Field

[0001] This invention belongs to the field of machine learning and relates to a method, apparatus, device, and medium for predicting the association between multiple non-coding RNAs and diseases. Background Technology

[0002] The human body contains many types of RNA, among which non-coding RNA (RNA), which does not encode proteins, plays a crucial role in gene expression and physiological changes, such as genomic imprinting, cell differentiation and variation, tumorigenesis, and disease development. Exploring the relationship between RNA and disease can effectively advance our understanding of the fundamental aspects of complex diseases and aid in the development of related drugs, and is currently one of the hottest research directions in the medical field. Early research methods primarily relied on biological experiments to determine associations. While these methods yielded highly accurate results and could sufficiently demonstrate the association between RNA and disease, the large number of biological experiments involved significant costs and lengthy experimental cycles. Subsequently, based on the large datasets obtained from early biological experiments, the biological research community introduced machine learning methods to predict the association between non-coding RNA and disease. Although this method reduced prediction accuracy to some extent, it offered high efficiency and greatly advanced our understanding of the relationship between the two.

[0003] Predicting the relationship between non-coding RNAs and diseases using traditional machine learning methods largely relies on algorithms based on tensor factorization, random forests, or matrix correlation. For example, a novel model called PADLMHOOI has been proposed to predict the potential association between diseases and lncRNA-miRNA pairs based on high-order orthogonal iteration. A new computational method based on an autoencoder (DFELMDA) and random forest ensemble learning has been proposed to predict miRNA-disease associations. Further research has developed the Laplace-Regulated Least Squares (LRLSLDA) method for lncRNA-disease associations. These methods effectively utilize the advantages of traditional machine learning; however, their prediction efficiency decreases significantly with increasing data volume, and they perform poorly in predicting deep associations.

[0004] Meanwhile, methods also utilize deep models for association prediction, employing relatively simple multilayer perceptrons, autoencoders, or neural networks built based on graph structures and self-attention mechanisms. For example, a miRNA-disease-gene heterogeneous information network was constructed, defining seven symmetric meta-paths based on different semantics. Then, the vector information carried by all nodes on the meta-path instances was extracted and aggregated to update the feature vector of the starting node. Finally, the miRNA and disease embedding feature vectors were used to calculate the prediction score. In the application of graph neural networks, a new method based on graph convolutional networks and convolutional neural networks, GCNLDA, has been proposed. This new framework based on graph convolutional networks and convolutional neural networks has been developed to learn the network and local representations of incRNA disease pairs. Similarly, there are currently some mainstream methods based on attention mechanisms. For instance, LDAformer, a new method for predicting incRNA disease associations based on topological feature extraction and a Transformer encoder, was proposed. It designs a topological feature extraction process to further obtain potential multi-hop topological path features in the adjacent matrix. Graph-BERT has also been proposed, which is based solely on attention mechanisms without any graph convolution or aggregation operators. GRAPH-BERT is trained using sampled unconnected subgraphs within the local context, rather than the entire large input graph. While these methods leverage deep networks to improve the efficiency of large-sample predictions and gain a deeper understanding of global information, they neglect the efficiency gains from local information, lack information extraction at different scales, and are not conducive to obtaining richer feature representations, thus reducing prediction accuracy to some extent.

[0005] Overall, existing miRNA-incRNA combination and disease association prediction schemes suffer from the following problems: First, existing methods focus on analyzing the association between single non-coding RNAs and diseases, while methods for analyzing the association between multiple non-coding RNAs and diseases are still lacking. Second, traditional experimental validation methods are time-consuming and labor-intensive, and due to the limitation of long experimental cycles, the obtained prediction information is often incomplete, failing to meet the needs of prediction with large numbers of samples. Finally, the prediction efficiency of traditional machine learning methods will significantly decrease with further increases in data volume, and because they cannot obtain deeper feature information, they perform poorly in predicting more complex associations; while deep learning-based schemes suffer from a certain degree of information loss due to insufficient capture of local information, and most methods rely on existing data for training, leading to a certain degree of data leakage in prediction tasks. Summary of the Invention

[0006] The purpose of this invention is to overcome the shortcomings of existing non-coding RNA and disease association prediction methods, which are mostly focused on the analysis of the association between a single non-coding RNA and a disease, and cannot achieve the analysis of the association between multiple non-coding RNAs and diseases, and have low prediction efficiency and the risk of data leakage. This invention provides a method, device, equipment and medium for predicting the association between multiple non-coding RNAs and diseases.

[0007] To achieve the above objectives, the present invention employs the following technical solution:

[0008] In a first aspect, the present invention provides a method for predicting the association between multiple non-coding RNAs and diseases, comprising:

[0009] Constructing a bioinformatics network among miRNAs, lncRNAs, and diseases;

[0010] Based on the bioinformatics network, global information of the first node and the second node are extracted using a graph-based Laplace transform and absolute role encoding method; where the first node and the second node are both nodes in the bioinformatics network.

[0011] Construct unconnected subgraphs of the first node and the second node, and based on the unconnected subgraphs of the first node and the second node, aggregate the neighbor node features of the first node and the neighbor node features of the second node to obtain the local information of the first node and the local information of the second node.

[0012] The feature representation of the first node is obtained by concatenating the global and local information of the first node; the feature representation of the second node is obtained by concatenating the global and local information of the second node.

[0013] Input the feature representations of the first node and the feature representations of the second node into the preset correlation prediction model to obtain the correlation prediction results of the first node and the second node.

[0014] Optionally, the step of extracting the global information of the first node and the global information of the second node based on the bioinformatics network, through a graph-based Laplace transform and absolute role encoding, includes:

[0015] Obtain the similarity matrix of the bioinformatics network, and obtain the similarity information of the first node and the second node based on the similarity matrix of the bioinformatics network;

[0016] Obtain the degree matrix and connection matrix of the graph structure of the bioinformatics network, and obtain the Laplacian matrix of the bioinformatics network based on the degree matrix and connection matrix of the graph structure. Then, obtain the Laplacian eigenvectors of the first node and the second node based on the Laplacian matrix.

[0017] Based on the bioinformatics network, the absolute role embedding codes of the first node and the second node are obtained by the Weisfeiler-Lehman algorithm.

[0018] By concatenating the similarity information, Laplacian feature vector, and absolute role embedding encoding of the first node, the global information of the first node is obtained; by concatenating the similarity information, Laplacian feature vector, and absolute role embedding encoding of the second node, the global information of the second node is obtained.

[0019] Optionally, obtaining the similarity matrix of the bioinformatics network and obtaining the similarity information of the first node and the similarity information of the second node based on the similarity matrix of the bioinformatics network includes:

[0020] Based on the directed acyclic graph model of diseases in the MESH database, for each disease d1, D1 represents the set of d1 and all its ancestor nodes. The calculation of any node d in D1... i Semantic contribution value of disease d1 :

[0021]

[0022] The disease d is calculated using the following formula. i and disease d j Semantic similarity between Sem_Sim(d) i d j ):

[0023]

[0024] Where D1 represents disease d i D2 is the set of all its ancestor nodes, where D2 represents disease d. j and the set of all its ancestor nodes, This represents the relationship between each disease in D1 and disease d. i The sum of semantic contribution values, This represents the relationship between each disease in D2 and disease d. j The sum of semantic contribution values;

[0025] The disease d is calculated using the following formula. i and disease d j Gaussian spectral kernel similarity between KD(d) i d j ):

[0026] KD(d i d j )=exp(-γ d ||IP(d i )-IP(d j )||2 )

[0027]

[0028] Where, γ d n is the normalization parameter. d IP(d) represents the number of diseases. i ) represents the d in the correlation matrix of a bioinformatics network. i The corresponding vector, IP(d) j ) represents the d in the correlation matrix of a bioinformatics network. j The corresponding vector;

[0029] The disease d is expressed by the following formula. i and disease d j The semantic similarity between the two is integrated with the Gaussian spectral kernel similarity to obtain the disease d. i and disease d j The similarity between Sim(d) i d j ):

[0030]

[0031] RNAr is obtained using the following formula. i and RNAr j Functional similarity between Fun_Sim(r) i r j ):

[0032]

[0033] Among them, RNAr i or RNAr j All are miRNAs or all are lncRNAs, disease d ip (1≤p≤n1) and d jq (1≤q≤n2) represent the diseases associated with RNAr1 and RNAr2, respectively, where n1 and n2 represent the number of diseases associated with RNAr1 and RNAr2, respectively.

[0034] According to disease d i and disease d j Similarity between them and RNAr i and RNAr j The functional similarity between them is used to obtain the similarity matrix of the bioinformatics network;

[0035] Obtain the row vector corresponding to the first node in the similarity matrix to get the similarity information of the first node;

[0036] Obtain the row vector corresponding to the second node in the similarity matrix to get the similarity information of the second node.

[0037] Optionally, obtaining the degree matrix and connection matrix of the graph structure of the bioinformatics network, obtaining the Laplace matrix of the bioinformatics network based on the degree matrix and connection matrix of the graph structure, and obtaining the Laplace eigenvectors of the first node and the second node based on the Laplace matrix includes:

[0038] The Laplace matrix L of the bioinformatics network is obtained by the following formula: L = DA; where D represents the degree matrix of the graph structure of the bioinformatics network, and A represents the connection matrix of the bioinformatics network.

[0039] The normalized Laplace moments are then normalized to obtain the normalized Laplace matrix L. sym :L sym =D -1 / 2 LD -1 / 2 =ID -1 / 2 AD -1 / 2 ;

[0040] Obtain the row vector corresponding to the first node in the normalized Laplacian matrix to obtain the Laplacian eigenvector of the first node; obtain the row vector corresponding to the second node in the normalized Laplacian matrix to obtain the Laplacian eigenvector of the second node.

[0041] Optionally, obtaining the absolute role embedding code of the first node and the absolute role embedding code of the second node using the Weisfeiler-Lehman algorithm based on the bioinformatics network includes:

[0042] For each node v j Obtain the characteristics of the set of nodes connected to it. Then, the features of the surrounding point set are updated to the current node using a hash injective function. Repeat this operation until convergence, such that each node v j The corresponding character features are embedded in WL(v) across the entire map. j );

[0043] Each node v is obtained using the following formula. j Absolute role embedding encoding

[0044]

[0045]

[0046] in, Index l is traversed from 0 to Using trigonometric functions on WL(v)j The encoding is performed, where N represents the total number of nodes in the bioinformatics network.

[0047] Optionally, the step of constructing the disconnected subgraphs of the first node and the second node, and based on the disconnected subgraphs of the first node and the second node, aggregating the neighbor node features of the first node and the neighbor node features of the second node to obtain the local information of the first node and the local information of the second node includes:

[0048] The node affinity matrix S of the bioinformatics network is obtained using the following formula: in, D represents the degree matrix of the graph structure of the biological information network, A represents the connection matrix of the biological information network, and the parameter α∈[0,1];

[0049] For node v i Define its context node set Where, θ i Nodes with a pre-defined intimacy threshold are not considered nodes v. i In the unconnected subgraph, nodes are selected based on the node affinity matrix of the bioinformatics network, choosing nodes with affinity to node v. i The nodes with the highest affinity are selected, and their corresponding vectors in the similarity matrix of the bioinformatics network are concatenated to obtain node v. i Local information, taking the first node and the second node as node v respectively. i By performing the above steps, we obtain the local information of the first node and the local information of the second node.

[0050] Optionally, the preset correlation prediction model is constructed in the following manner:

[0051] The encoder of the Transformer model is connected to the multilayer perceptron network to obtain the initial correlation prediction model. The initial correlation prediction model is trained based on the training data to obtain the preset correlation prediction model. The encoder part of the Transformer model is used to input the feature representations of the first node and the feature representations of the second node, and the multilayer perceptron network is used to output the correlation prediction results of the first node and the second node.

[0052] In a second aspect, the present invention provides a multi-noncoding RNA and disease association prediction system, comprising:

[0053] The network building module is used to construct a bioinformatics network among miRNAs, lncRNAs, and diseases.

[0054] The global module is used to extract global information of the first node and the second node based on the bioinformatics network by using a graph-based Laplace transform and absolute role encoding; where the first node and the second node are both nodes in the bioinformatics network.

[0055] The local module is used to construct the unconnected subgraph of the first node and the unconnected subgraph of the second node, and based on the unconnected subgraph of the first node and the unconnected subgraph of the second node, it aggregates the neighbor node features of the first node and the neighbor node features of the second node to obtain the local information of the first node and the local information of the second node.

[0056] The concatenation module is used to concatenate the global and local information of the first node to obtain the feature representation of the first node, and to concatenate the global and local information of the second node to obtain the feature representation of the second node.

[0057] The prediction module is used to input the feature representations of the first node and the feature representations of the second node into a preset correlation prediction model to obtain the correlation prediction results of the first node and the second node.

[0058] In a third aspect, the present invention provides a computer device including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the steps of the above-described method for predicting multiple non-coding RNAs and disease associations.

[0059] In a fourth aspect, the present invention provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the steps of the above-described method for predicting the association between multiple non-coding RNAs and diseases.

[0060] Compared with the prior art, the present invention has the following beneficial effects:

[0061] This invention presents a method for predicting the association between multiple non-coding RNAs and diseases. In terms of feature extraction, it acquires multi-scale information that integrates global and local information. First, it uses the Laplace transform based on a graph structure and the absolute role encoding of each node to extract features, fully acquiring global information based on the whole graph structure. Then, it uses the neighbor node information aggregation method for a certain node to obtain a local information representation about that node, and integrates it with the previous global information to obtain a multi-scale feature representation, acquiring richer node information and effectively improving the accuracy of the prediction results. In terms of data, it constructs a bioinformatics network among miRNAs, lncRNAs, and diseases to realize the association analysis between multiple non-coding RNAs and diseases. It achieves superior performance compared to existing technologies in both data leakage and no data leakage scenarios. Attached Figure Description

[0062] Figure 1 This is a flowchart of the method for predicting multiple non-coding RNAs and disease associations according to an embodiment of the present invention.

[0063] Figure 2 This is a schematic diagram illustrating the principle of the multiple non-coding RNA and disease association prediction method according to an embodiment of the present invention.

[0064] Figure 3 This is a comparison chart of ROC curves under different models in embodiments of the present invention.

[0065] Figure 4 This is a comparison chart of PR curves under different models in embodiments of the present invention.

[0066] Figure 5 This is a block diagram of the multi-noncoding RNA and disease association prediction system according to an embodiment of the present invention. Detailed Implementation

[0067] To enable those skilled in the art to better understand the present invention, the technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort should fall within the scope of protection of the present invention.

[0068] It should be noted that the terms "first," "second," etc., in the specification, claims, and accompanying drawings of this invention are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments of the invention described herein can be implemented in orders other than those illustrated or described herein. Furthermore, the terms "comprising" and "having," and any variations thereof, are intended to cover a non-exclusive inclusion; for example, a process, method, system, product, or apparatus that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or apparatus.

[0069] The present invention will now be described in further detail with reference to the accompanying drawings:

[0070] See Figure 1 and 2 In one embodiment of the present invention, a method for predicting the association between multiple non-coding RNAs and diseases is provided, comprising the following steps:

[0071] S1: Construct a bioinformatics network among miRNAs, lncRNAs, and diseases. During construction, miRNAs, lncRNAs, and diseases are all treated as nodes, and the relationships between them are represented as edges.

[0072] S2: Based on the bioinformatics network, global information of the first node and the second node are extracted using a graph-based Laplace transform and absolute role encoding method. Both the first node and the second node are nodes within the bioinformatics network.

[0073] S3: Construct the unconnected subgraph of the first node and the unconnected subgraph of the second node, and based on the unconnected subgraph of the first node and the unconnected subgraph of the second node, aggregate the neighbor node features of the first node and the neighbor node features of the second node to obtain the local information of the first node and the local information of the second node.

[0074] S4: Concatenate the global and local information of the first node to obtain the feature representation of the first node, and concatenate the global and local information of the second node to obtain the feature representation of the second node.

[0075] S5: Input the feature representations of the first node and the feature representations of the second node into the preset correlation prediction model to obtain the correlation prediction results of the first node and the second node.

[0076] This invention presents a method for predicting the association between multiple non-coding RNAs and diseases. In terms of feature extraction, it acquires multi-scale information that integrates global and local information. First, it uses the Laplace transform based on a graph structure and the absolute role encoding of each node to extract features, fully acquiring global information based on the whole graph structure. Then, it uses the neighbor node information aggregation method for a certain node to obtain a local information representation about that node, and integrates it with the previous global information to obtain a multi-scale feature representation, acquiring richer node information and effectively improving the accuracy of the prediction results. In terms of data, it constructs a bioinformatics network among miRNAs, lncRNAs, and diseases to realize the association analysis between multiple non-coding RNAs and diseases. It achieves superior performance compared to existing technologies in both data leakage and no data leakage scenarios.

[0077] This invention relates to a method for predicting disease associations between multiple non-coding RNAs, specifically a method for predicting disease associations based on a combination of miRNAs and lncRNAs using multi-scale information fusion. To make the technical solution and implementation process of this invention clearer and more understandable, the specific scheme of the overall method is described below. These implementation schemes are only used to explain the technical principles of this invention, but are not limited to the application scope of this invention.

[0078] In one possible implementation, step S1 specifically involves: acquiring official datasets of miRNAs, lncRNAs, and diseases, and constructing a bioinformatics molecular network. in Let ε represent the set of nodes in the network, and let ε represent the set of connected edges in the network.

[0079] Specifically, we collected official datasets on miRNAs, lncRNAs, and diseases from various sources, such as miRCancer, LncRNADisease, and RNADisease. Based on the known associations between these datasets, we removed redundant data and constructed a multi-source biomolecular network. in Let represent the set of nodes in the network, and ε represent the set of edges in the network. In this example, there are 464 miRNAs, 56 lncRNAs, and 121 diseases, totaling 641 nodes and 7322 edges. The connection matrix corresponding to this network is A∈R. N×N (N=641). When a certain disease node d i With a certain miRNA node m i When a relationship that has been experimentally verified exists, then the element A(d) in the connection matrix... i m i ) = 1; Similarly, if no verified relation exists, then the corresponding element A(d) in the connection matrix is ​​1. i m i ) = 0.

[0080] In one possible implementation, the step of extracting global information of the first node and the second node based on the bioinformatics network and using a graph-based Laplacian transform and absolute role encoding includes: obtaining the similarity matrix of the bioinformatics network, and obtaining the similarity information of the first node and the second node based on the similarity matrix of the bioinformatics network; obtaining the degree matrix and connection matrix of the graph structure of the bioinformatics network, and obtaining the Laplacian matrix of the bioinformatics network based on the degree matrix and connection matrix of the graph structure, and obtaining the Laplacian eigenvectors of the first node and the second node based on the Laplacian eigenvectors; obtaining the absolute role embedding encoding of the first node and the second node using the Weisfeiler-Lehman algorithm based on the bioinformatics network; concatenating the similarity information, Laplacian eigenvectors, and absolute role embedding encoding of the first node to obtain the global information of the first node; and concatenating the similarity information, Laplacian eigenvectors, and absolute role embedding encoding of the second node to obtain the global information of the second node.

[0081] In one possible implementation, obtaining the similarity matrix of the bioinformatics network and obtaining the similarity information of the first node and the similarity information of the second node based on the similarity matrix of the bioinformatics network includes: based on the directed acyclic graph model of diseases in the MESH database, for each disease d1, D1 represents the set of d1 and all its ancestor nodes, calculating the similarity information of any node d in D1. i Semantic contribution value of disease d1

[0082]

[0083] The disease d is calculated using the following formula. i and disease d j Semantic similarity between Sem_Sim(d) i d j ):

[0084]

[0085] Where D1 represents disease d i D2 is the set of all its ancestor nodes, where D2 represents disease d. j and the set of all its ancestor nodes, This represents the relationship between each disease in D1 and disease d. i The sum of semantic contribution values, This represents the relationship between each disease in D2 and disease d. j The sum of semantic contribution values.

[0086] The disease d is calculated using the following formula. i and disease d j Gaussian spectral kernel similarity between KD(d) i d j ):

[0087] KD(d i d j )=exp(-γ d ||IP(d i )-IP(d j )|| 2 )

[0088]

[0089] Where, γ d n is the normalization parameter. d IP(d) represents the number of diseases. i ) represents the d in the correlation matrix of a bioinformatics network. i The corresponding vector, IP(d) j ) represents the d in the correlation matrix of a bioinformatics network. jThe corresponding vector.

[0090] The disease d is expressed by the following formula. i and disease d j The semantic similarity between the two is integrated with the Gaussian spectral kernel similarity to obtain the disease d. i and disease d j The similarity between Sim(d) i d j ):

[0091]

[0092] RNAr is obtained using the following formula. i and RNAr j Functional similarity between Fun_Sim(r) i r j ):

[0093]

[0094] Among them, RNAr i or RNAr j All are miRNAs or all are lncRNAs, disease d ip (1≤p≤n1) and d jq (1≤q≤n2) represent the diseases associated with RNAr1 and RNAr2, respectively, and n1 and n2 represent the number of diseases associated with RNAr1 and RNAr2, respectively.

[0095] According to disease d i and disease d j Similarity between them and RNAr i and RNAr j The functional similarity between them is used to obtain the similarity matrix of the bioinformatics network.

[0096] Obtain the row vector corresponding to the first node in the similarity matrix to get the similarity information of the first node.

[0097] Obtain the row vector corresponding to the second node in the similarity matrix to get the similarity information of the second node.

[0098] Specifically, firstly, based on the directed acyclic graph (DAG) model of diseases in the MESH database, the semantic similarity matrix between diseases is calculated. After obtaining the semantic similarity between diseases, Gaussian kernel similarity between diseases is obtained through the connection relationships between diseases and miRNAs, and between diseases and lncRNAs. Then, the semantic similarity and Gaussian kernel similarity are integrated to obtain the final similarity between diseases. Finally, the functional similarity between miRNAs / lncRNAs is obtained using the disease semantic similarity and the disease-RNA association matrix.

[0099] After obtaining the semantic similarity, Gaussian kernel similarity, functional similarity between miRNAs, and functional similarity between lncRNAs, the constructed similarity information is added to the original connection matrix A to obtain the final similarity matrix M∈R. N×N .

[0100] In one possible implementation, obtaining the degree matrix and connection matrix of the graph structure of the bioinformatics network, obtaining the Laplace matrix of the bioinformatics network based on the degree matrix and connection matrix, and obtaining the Laplace eigenvectors of the first node and the second node based on the Laplace matrix includes: obtaining the Laplace matrix L of the bioinformatics network using the following formula: L = DA; where D represents the degree matrix of the graph structure of the bioinformatics network, and A represents the connection matrix of the bioinformatics network; and normalizing the Laplace moments to obtain the normalized Laplace matrix L. sym :L sym =D -1 / 2 LD -1 / 2 =ID -1 / 2 AD -1 / 2 Obtain the row vector corresponding to the first node in the normalized Laplacian matrix to obtain the Laplacian eigenvector of the first node; obtain the row vector corresponding to the second node in the normalized Laplacian matrix to obtain the Laplacian eigenvector of the second node.

[0101] Specifically, based on the undirected graph structure, its Laplacian matrix L = DA is obtained, and then normalized to obtain L. sym =D -1 / 2 LD -1 / 2 =ID -1 / 2 AD -1 / 2 Where D represents the degree matrix of the graph structure, and A represents the original connection matrix. Each node v i The corresponding row vector This is the Laplacian eigenvector of the node, and the vector dimension is the same as the number of nodes in the graph.

[0102] In one possible implementation, obtaining the absolute role embedding code of the first node and the absolute role embedding code of the second node using the Weisfeiler-Lehman algorithm according to the bioinformatics network includes: for each node vj Obtain the characteristics of the set of nodes connected to it. Then, the features of the surrounding point set are updated to the current node using a hash injective function. Repeat this operation until convergence, such that each node v j The corresponding character features are embedded in WL(v) across the entire map. j Then, the value of each node v is obtained through the following formula. j Absolute role embedding encoding

[0103]

[0104] in, Index l is traversed from 0 to Using trigonometric functions on WL(v) j The encoding is performed, where N represents the total number of nodes in the bioinformatics network.

[0105] In one possible implementation, constructing the disconnected subgraphs of the first node and the second node, and based on these disconnected subgraphs, aggregating the neighbor node features of the first node and the neighbor node features of the second node to obtain the local information of the first node and the second node, includes obtaining the node affinity matrix S of the bioinformatics network using the following formula: in, D represents the degree matrix of the graph structure of the bioinformatics network, A represents the connection matrix of the bioinformatics network, and the parameter α∈[0,1]; for node v i Define its context node set Where, θ i Nodes with a pre-defined intimacy threshold are not considered nodes v. i In the unconnected subgraph, nodes are selected based on the node affinity matrix of the bioinformatics network, choosing nodes with affinity to node v. i The nodes with the highest affinity are selected, and their corresponding vectors in the similarity matrix of the bioinformatics network are concatenated to obtain node v. i Local information, taking the first node and the second node as node v respectively. i By performing the above steps, we obtain the local information of the first node and the local information of the second node.

[0106] Specifically, the node affinity matrix is ​​calculated using the original connection matrix. The affinity matrix is ​​defined as follows: in Let v represent the row-normalized connection matrix, with parameter α∈[0,1] (defined as 0.5 in this embodiment). For a given node v iDefine its context node set Where θ i A preset intimacy threshold is set; nodes below this threshold are not considered v. i The nodes in the unconnected subgraph. Then select the 5 nodes with the highest affinity and v. i Constructing unconnected subgraphs and The vectors corresponding to the 5 nodes in the similarity matrix M above. splicing together yields node v i The local feature matrix.

[0107] In one possible implementation, the global information of the first node is obtained by concatenating the similarity information, Laplacian feature vector, and absolute role embedding encoding of the first node; the global information of the second node is obtained by concatenating the similarity information, Laplacian feature vector, and absolute role embedding encoding of the second node, including:

[0108] node v i The similarity information corresponding to the similarity matrix M The corresponding eigenvalue in the Laplacian matrix and absolute role coding The global feature matrix GM is obtained by concatenating the three feature vectors. i That is, global information. This involves extracting the similarity features of the five neighboring nodes obtained from the local information extraction. The local feature matrix LM is obtained by concatenation. i That is, local information. Ultimately, GM... i With LM i splicing together yields node v i The input feature matrix M i =[GM i LM i ] T That is, feature representation.

[0109] In one possible implementation, the preset correlation prediction model is constructed as follows: the encoder of the Transformer model is connected to a multilayer perceptron network to obtain an initial correlation prediction model; the initial correlation prediction model is trained based on training data to obtain the preset correlation prediction model; wherein, the encoder part of the Transformer model is used to input the feature representation of the first node and the feature representation of the second node, and the multilayer perceptron network is used to output the correlation prediction results of the first node and the second node.

[0110] In this embodiment, the feature representations of two nodes are used as input to the encoder part of the Transformer model. Multi-head attention and multi-layer self-attention encoding are used to obtain the feature encoding, and the dimensionality of the attention part is reduced. The final output is expanded into a one-dimensional form and processed using a multi-layer perceptron network to output the prediction result. Five-fold cross-validation is used to train the model, and regularization and Dropout are used to prevent overfitting. Finally, the obtained prediction results are compared with other datasets, and the AUC and AUPR values ​​during training are obtained.

[0111] Specifically, the two nodes v i and v j The feature representation is used as the input to the encoder part of the Transformer model, i.e., M = [M i M j ] T Feature encoding is obtained using a multi-head attention mechanism and a multi-layer self-attention encoding method. The feature dimension of the attention part is reduced to 512. The self-attention calculation is as follows:

[0112]

[0113]

[0114] Each attention head uses three different matrices to multiply the input X, transforming it into Q. i K i and V i :

[0115] Q i =XW i Q K i =XW i K V i =XW i V

[0116] Where 1≤i≤n head And W i Q W i K and All are learnable matrices; different output heads are concatenated using matrices. After performing a linear transformation, the final output header MHA(X) is calculated as follows:

[0117]

[0118] head i=Attention(Q) i K i V i )

[0119] Then, MHA(X) is added to the original input X and normalized. The result is used as the new input to the feedforward network FFN, calculated as follows:

[0120] X = Layer_Norm1(X + MHA(X))

[0121] FFN(X)=max(0,XW1+b i W2+b2

[0122] The output FFN(X) of the feedforward network is added to the input X, and the result is used as the input to the second normalization layer, calculated as follows:

[0123] Encoder_Layer(X)=Layer_Norm2(X+FFN(X))

[0124] The final output Encoder_Layer(X) is the single-layer encoded output.

[0125] In this embodiment, the number of encoding layers is set to 8. After the above encoding process, the final encoded output is X. encoded After being expanded into a one-dimensional tensor, the final output is obtained through a prediction layer consisting of fully connected layers, i.e., a multilayer perceptron. The loss function is defined using cross-entropy and calculated as follows:

[0126] pred = Pred_Layer(flatten(X) encoded ))

[0127] loss=-∑[ylog(pred)+(1-y)log(1-pred)]

[0128] Here, `pred` represents the model's predicted value, and `y` represents the actual value. A three-layer MLP structure is used in the prediction layer `Pred_Layer`, progressively reducing the expanded encoding dimension to 1024, 256, and 1 dimension. When the true value `y` is 1, the larger the predicted value `pred`, the smaller the loss function; conversely, the larger the predicted value `pred`, the larger the loss function.

[0129] In one possible implementation, the experiment was conducted under the following conditions: single-card operation on a TITAN GPU server with Ubuntu 18.04 operating system, and a single training round (epochs = 30) of approximately 20 minutes.

[0130] The experiment used a five-fold cross-validation approach, with evaluation metrics including AUC, AUPR, and ACC. AUC was obtained by calculating the area under the ROC curve, PR was obtained by calculating the coordinates under the PR curve, and ACC was obtained by comparing the ratio of correctly predicted results to the total predicted samples.

[0131] See Figure 3 and 4 The experiment yielded ROC and PR curves, and the results show a comparison between the method of this invention and other methods. The disease-related LncRNA-miRNA pair prediction method based on high-order orthogonal iteration (PADLMHOOI) is a traditional machine learning method. The convolutional neural network-based method (CNNMDA) and the autoencoder-based random forest ensemble learning method (DFELMDA) both utilize general neural networks for association prediction. Furthermore, the multi-relationship graph convolutional neural network based on a high-order graph structure of miRNA-gene-disease (HGCNMDA) utilizes a heterogeneous network composed of different molecules, combined with a graph neural network for association prediction. LDAformer uses a self-attention model for predictive analysis. This comparison covers a wide range of methods, and the experimental results effectively confirm the accuracy of the prediction results of this method. It also shows that the traditional machine learning method (PADLMHOOI) performs poorly with large datasets, and that as the model complexity and structure increase (from CNNMDA to BertNDA), it exhibits better performance in uncovering potential connections.

[0132] The following are embodiments of the apparatus of the present invention, which can be used to execute embodiments of the method of the present invention. For details not disclosed in the apparatus embodiments, please refer to the embodiments of the method of the present invention.

[0133] See Figure 5 In another embodiment of the present invention, a multiple non-coding RNA and disease association prediction system is provided, which can be used to implement the above-mentioned multiple non-coding RNA and disease association prediction method. Specifically, the multiple non-coding RNA and disease association prediction system includes a network construction module, a global module, a local module, a splicing module, and a prediction module.

[0134] The system comprises the following modules: a network construction module for building a bioinformatics network among miRNAs, lncRNAs, and diseases; a global module for extracting global information of the first and second nodes based on the bioinformatics network using a graph-based Laplace transform and absolute role encoding; a local module for constructing unconnected subgraphs of the first and second nodes, and aggregating the features of the neighboring nodes of the first and second nodes to obtain local information of the first and second nodes; a splicing module for concatenating the global and local information of the first node to obtain a feature representation of the first node, and concatenating the global and local information of the second node to obtain a feature representation of the second node; and a prediction module for inputting the feature representations of the first and second nodes into a preset correlation prediction model to obtain the correlation prediction results of the first and second nodes.

[0135] All relevant content of each step involved in the aforementioned embodiments of the multiple non-coding RNA and disease association prediction method can be referenced in the functional description of the corresponding functional module of the multiple non-coding RNA and disease association prediction system in the embodiments of the present invention, and will not be repeated here.

[0136] The module division in this embodiment of the invention is illustrative and represents only one logical functional division. In actual implementation, other division methods may be used. Furthermore, the functional modules in the various embodiments of the invention can be integrated into a single processor, exist as separate physical entities, or be integrated into a single module. The integrated modules described above can be implemented in hardware or as software functional modules.

[0137] In another embodiment of the present invention, a computer device is provided, comprising a processor and a memory. The memory stores a computer program, which includes program instructions. The processor executes the program instructions stored in the computer storage medium. The processor may be a Central Processing Unit (CPU), or other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. It is the computing and control core of the terminal, suitable for implementing one or more instructions, specifically suitable for loading and executing one or more instructions in the computer storage medium to achieve a corresponding method flow or corresponding function. The processor described in this embodiment of the present invention can be used for the operation of multiple non-coding RNA and disease association prediction methods.

[0138] In another embodiment of the present invention, a storage medium is provided, specifically a computer-readable storage medium (Memory), which is a memory device in a computer device used to store programs and data. It is understood that the computer-readable storage medium here can include both the built-in storage medium in the computer device and extended storage media supported by the computer device. The computer-readable storage medium provides storage space that stores the terminal's operating system. Furthermore, the storage space also stores one or more instructions suitable for loading and execution by a processor. These instructions can be one or more computer programs (including program code). It should be noted that the computer-readable storage medium here can be high-speed RAM or non-volatile memory, such as at least one disk storage device. The processor can load and execute one or more instructions stored in the computer-readable storage medium to implement the corresponding steps of the multiple non-coding RNA and disease association prediction method in the above embodiments.

[0139] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, systems, or computer program products. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention can take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0140] This invention is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart illustrations and / or block diagrams. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0141] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0142] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0143] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit it. Although the present invention has been described in detail with reference to the above embodiments, those skilled in the art should understand that modifications or equivalent substitutions can still be made to the specific implementation of the present invention. Any modifications or equivalent substitutions that do not depart from the spirit and scope of the present invention should be covered within the scope of protection of the claims of the present invention.

Claims

1. A method for predicting the association between multiple non-coding RNAs and diseases, characterized in that, include: Constructing a bioinformatics network among miRNAs, lncRNAs, and diseases; Based on the bioinformatics network, global information of the first node and the second node are extracted using a graph-based Laplace transform and absolute role encoding method; where the first node and the second node are both nodes in the bioinformatics network. Construct unconnected subgraphs of the first node and the second node, and based on the unconnected subgraphs of the first node and the second node, aggregate the neighbor node features of the first node and the neighbor node features of the second node to obtain the local information of the first node and the local information of the second node. The feature representation of the first node is obtained by concatenating the global and local information of the first node; the feature representation of the second node is obtained by concatenating the global and local information of the second node. Input the feature representations of the first node and the feature representations of the second node into the preset correlation prediction model to obtain the correlation prediction results of the first node and the second node. The step of extracting global information of the first node and the second node based on the bioinformatics network, using a graph-based Laplace transform and absolute role encoding method, includes: Obtain the similarity matrix of the bioinformatics network, and obtain the similarity information of the first node and the second node based on the similarity matrix of the bioinformatics network; Obtain the degree matrix and connection matrix of the graph structure of the bioinformatics network, and obtain the Laplacian matrix of the bioinformatics network based on the degree matrix and connection matrix of the graph structure. Then, obtain the Laplacian eigenvectors of the first node and the second node based on the Laplacian matrix. Based on the bioinformatics network, the absolute role embedding codes of the first node and the second node are obtained by the Weisfeiler-Lehman algorithm. By concatenating the similarity information, Laplacian feature vector, and absolute role embedding encoding of the first node, the global information of the first node is obtained; by concatenating the similarity information, Laplacian feature vector, and absolute role embedding encoding of the second node, the global information of the second node is obtained. The preset correlation prediction model is constructed in the following way: The encoder of the Transformer model is connected to the multilayer perceptron network to obtain the initial correlation prediction model. The initial correlation prediction model is trained based on the training data to obtain the preset correlation prediction model. The encoder part of the Transformer model is used to input the feature representations of the first node and the feature representations of the second node, and the multilayer perceptron network is used to output the correlation prediction results of the first node and the second node.

2. The method for predicting the association between multiple non-coding RNAs and diseases according to claim 1, characterized in that, The step of obtaining the similarity matrix of the bioinformatics network and obtaining the similarity information of the first node and the similarity information of the second node based on the similarity matrix of the bioinformatics network includes: Based on the directed acyclic graph model of diseases in the MESH database, for each disease , express Given the set of all its ancestor nodes, calculate any node With disease semantic contribution value : The disease is calculated using the following formula. and disease semantic similarity between : in, Indicates disease and the set of all its ancestor nodes, Indicates disease and the set of all its ancestor nodes, express Various diseases and diseases The sum of semantic contribution values, express Various diseases and diseases The sum of semantic contribution values; The disease is calculated using the following formula. and disease Gaussian spectral kernel similarity between : in, For normalization parameters, Indicates the number of diseases. In the association matrix representing bioinformatics networks The corresponding vector, In the association matrix representing bioinformatics networks The corresponding vector; The disease is expressed by the following formula. and disease The semantic similarity between the two is integrated with the Gaussian spectral kernel similarity to obtain the disease. and disease Similarity between : RNA is obtained using the following formula. and RNA Functional similarity between : Among them, RNA or RNA All are miRNAs or all are lncRNAs, disease and They represent RNA respectively and RNA Related diseases, and Respectively represent and and RNA The number of related diseases; According to the disease and disease Similarity between them and RNA and RNA The functional similarity between them is used to obtain the similarity matrix of the bioinformatics network; Obtain the row vector corresponding to the first node in the similarity matrix to get the similarity information of the first node; Obtain the row vector corresponding to the second node in the similarity matrix to get the similarity information of the second node.

3. The method for predicting the association between multiple non-coding RNAs and diseases according to claim 1, characterized in that, The process of obtaining the degree matrix and connection matrix of the graph structure of the bioinformatics network, obtaining the Laplace matrix of the bioinformatics network based on the degree matrix and connection matrix, and obtaining the Laplace eigenvectors of the first node and the second node based on the Laplace matrix includes: The Laplace matrix of the bioinformatics network is obtained using the following formula. : ;in, D The degree matrix representing the graph structure of a bioinformatics network. A Represents the connection matrix of a bioinformatics network; The normalized Laplacian matrix is ​​obtained by normalizing the Laplacian moments. : ; Obtain the row vector corresponding to the first node in the normalized Laplacian matrix to obtain the Laplacian eigenvector of the first node; obtain the row vector corresponding to the second node in the normalized Laplacian matrix to obtain the Laplacian eigenvector of the second node.

4. The method for predicting the association between multiple non-coding RNAs and diseases according to claim 1, characterized in that, The step of obtaining the absolute role embedding code of the first node and the absolute role embedding code of the second node using the Weisfeiler-Lehman algorithm based on the bioinformatics network includes: For each node Obtain the characteristics of the set of nodes connected to it. Then, the features of the surrounding point set are updated to the current node using a hash injective function. Repeat this operation until convergence, so that each node... Corresponding character features are embedded throughout the entire image. ; Each node is obtained using the following formula. Absolute role embedding encoding : in, ,index Traversing from 0 to and use trigonometric functions to Encode and represent. N This represents the total number of nodes in a bioinformatics network.

5. The method for predicting the association between multiple non-coding RNAs and diseases according to claim 1, characterized in that, The process of constructing an unconnected subgraph of the first node and an unconnected subgraph of the second node, and aggregating the neighbor node features of the first node and the neighbor node features of the second node based on the unconnected subgraphs of the first node and the second node to obtain the local information of the first node and the local information of the second node includes: The node affinity matrix of the bioinformatics network is obtained using the following formula. : ;in, , D The degree matrix representing the graph structure of a bioinformatics network. A The connection matrix of a bioinformatics network, with parameters ; For nodes Define its context node set ;in, Nodes with a pre-defined intimacy threshold are not considered nodes. In the unconnected subgraph, nodes are selected based on the node affinity matrix of the bioinformatics network. The nodes with the highest affinity are selected, and their corresponding vectors in the similarity matrix of the bioinformatics network are concatenated to obtain the node. Local information, treating the first node and the second node as nodes respectively. By performing the above steps, we obtain the local information of the first node and the local information of the second node.

6. A system for predicting multiple non-coding RNAs and disease associations based on the method for predicting multiple non-coding RNAs and diseases according to claim 1, characterized in that, include: The network building module is used to construct a bioinformatics network among miRNAs, lncRNAs, and diseases. The global module is used to extract global information of the first node and the second node based on the bioinformatics network by using a graph-based Laplace transform and absolute role encoding; where the first node and the second node are both nodes in the bioinformatics network. The local module is used to construct the unconnected subgraph of the first node and the unconnected subgraph of the second node, and based on the unconnected subgraph of the first node and the unconnected subgraph of the second node, it aggregates the neighbor node features of the first node and the neighbor node features of the second node to obtain the local information of the first node and the local information of the second node. The concatenation module is used to concatenate the global and local information of the first node to obtain the feature representation of the first node, and to concatenate the global and local information of the second node to obtain the feature representation of the second node. The prediction module is used to input the feature representations of the first node and the feature representations of the second node into a preset correlation prediction model to obtain the correlation prediction results of the first node and the second node.

7. A computer device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the computer program, it implements the steps of the multiple non-coding RNA and disease association prediction method as described in any one of claims 1 to 5.

8. A computer-readable storage medium storing a computer program, characterized in that, When the computer program is executed by a processor, it implements the steps of the multiple non-coding RNA and disease association prediction method as described in any one of claims 1 to 5.

Images